ZhAgent/apps/models_provider/impl/xinference_model_provider/model/stt.py

import io
from typing import Dict

from openai import OpenAI

from common.config.tokenizer_manage_config import TokenizerManage
from models_provider.base_model_provider import MaxKBBaseModel
from models_provider.impl.base_stt import BaseSpeechToText


def custom_get_token_ids(text: str):
    tokenizer = TokenizerManage.get_tokenizer()
    return tokenizer.encode(text)


class XInferenceSpeechToText(MaxKBBaseModel, BaseSpeechToText):
    api_base: str
    api_key: str
    model: str
    params: dict

    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self.api_key = kwargs.get('api_key')
        self.api_base = kwargs.get('api_base')
        self.model = kwargs.get('model')
        self.params = kwargs.get('params')

    @staticmethod
    def is_cache_model():
        return False

    @staticmethod
    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
        optional_params = {}
        if 'max_tokens' in model_kwargs and model_kwargs['max_tokens'] is not None:
            optional_params['max_tokens'] = model_kwargs['max_tokens']
        if 'temperature' in model_kwargs and model_kwargs['temperature'] is not None:
            optional_params['temperature'] = model_kwargs['temperature']
        return XInferenceSpeechToText(
            model=model_name,
            api_base=model_credential.get('api_base'),
            api_key=model_credential.get('api_key'),
            params=model_kwargs,
            **optional_params,
        )

    def check_auth(self):
        client = OpenAI(
            base_url=self.api_base,
            api_key=self.api_key
        )
        response_list = client.models.with_raw_response.list()
        # print(response_list)

    def speech_to_text(self, audio_file):
        client = OpenAI(
            base_url=self.api_base,
            api_key=self.api_key
        )
        audio_data = audio_file.read()
        buffer = io.BytesIO(audio_data)
        buffer.name = "file.mp3"  # this is the important line

        filter_params = {k: v for k, v in self.params.items() if k not in {'model_id', 'use_local', 'streaming'}}
        transcription_params = {
            'model': self.model,
            'file': buffer,
            'language': 'zh',
            **filter_params
        }

        res = client.audio.transcriptions.create(**transcription_params)
        return res.text